Pendekatan Transfer Learning dan SMOTE untuk Klasifikasi Kanker Kulit pada Imbalanced Dataset

Abstract

Skin cancer is one of the most commonly diagnosed cancers worldwide, with the incidence increasing every year. While early detection is a key factor in reducing skin cancer mortality, conventional methods such as biopsy have limitations in terms of cost and invasiveness. This research applies a deep learning based approach for skin cancer classification with Convolutional Neural Networks (CNN) model using transfer learning method. 3 CNN architectures namely MobileNetV2, EfficientNetB0, and DenseNet121 are used to evaluate the performance of the model in detecting skin cancer. One of the main challenges in this research is the imbalanced dataset, which can cause bias in classification. The Synthetic Minority Over-Sampling Technique (SMOTE) was applied to improve the representation of minority classes. The dataset used comes from Kaggle and consists of 2,357 images classified into 9 skin cancer categories. The results show that the transfer learning method combined with SMOTE can significantly improve the accuracy of the model, especially in detecting classes with a smaller number of samples. The evaluation was conducted using accuracy, precision, recall, and f1-score metrics. This research is expected to contribute to the development of an artificial intelligence-based skin cancer detection system that is more accurate, efficient, and can be used as a tool for medical personnel in early diagnosis of skin cancer.